A Framework for Landslide Susceptibility Mapping for Minnesota Using Multi-Layer Perceptron Neural Network

A landslide is a geo-hazard that occurs due to the dislocation of soil or rock mass from the parent mass in the direction of gravitational force. Landslide events, often caused by a combination of natural and human-made factors, result in loss of lives and damage to property. In Minnesota, Wisconsin, and Michigan, approximately $6.7 million are spent annually on infrastructural repairs due to slope failures and landslides. Moreover, these failures cause major traffic delays, disrupt local businesses, and pose life-threatening risks to the nearby communities. Identification of landslide-susceptible zones can help improve urban management and prioritize the areas that need immediate remediation. Many factors can affect the trigger of a landslide event. The relationship between the landslide and its influencing factors is nonlinear. Artificial neural networks, mimicking the brain structure, offer a powerful tool for establishing and studying the non-linearity between variables. In the present study, a framework was demonstrated considering seven landslide influencing factors for modeling of the landslide susceptibility (LS) map for Minnesota. The LS map was developed using a multi-layer perceptron (MLP) neural network. In the current study, 7,700 landslide events in Minnesota and 7,707 non-landslide events were captured from various databases and considered for the development of the MLP model. The LS map for the study area was developed using the MLP model, which uses 70% and 30% of the dataset as the training and validation datasets, respectively. The accuracy obtained using the validation dataset was around 94.85%, indicating the “excellent”, match category. The developed LS map will be very helpful in guiding the state agencies to a better understanding of slope stability and plan for mitigation strategies.